Most NBFCs evaluating a debt collection platform spend 70% of the time looking at the wrong things - dashboard polish, the number of channels supported, the slickness of the demo. Those are table stakes.

The five questions below are the ones that actually separate a real debt collection platform from collections software with a new label. If a vendor can't answer all five clearly, the rest of the evaluation doesn't matter.

Question 1: Does it intervene before the EMI misses - or only after?

Most "platforms" only have workflows for accounts that are already overdue. Day 1 DPD. Day 7. Day 30. By the time those workflows fire, the highest-recovery-probability window - the 15 days before the EMI misses - is already closed.

A real debt collection platform runs an SMA-0 layer: predictive default scoring on every account, every day, with automated outreach in the pre-due window. Recovery probability in this window often sits above 80%. Cost of intervention: a WhatsApp message.

Ask the vendor: "Show me the workflow that fires for an account that is not yet overdue." If they don't have one, they don't have a platform. They have a dialer.

Question 2: Can it route accounts intelligently - or does it just create queues?

Watch the demo carefully. When the vendor shows account assignment, are accounts being routed to agents based on actual intelligence - risk tier, borrower channel preference, communication history - or are they being routed by DPD bucket and dropped into a queue?

Queue-based assignment is not routing. It is sorting. Every agent gets the same kind of account, in the same kind of state, treated the same way. The 24-year-old who responds to WhatsApp gets the same workflow as the 47-year-old who only answers calls. The first-time misser gets the same intensity as the strategic defaulter.

Real routing is per-account, per-borrower, per-channel. Same DPD bucket, two different accounts, two different treatment paths. That is the difference between manual triage and intelligence.

Question 3: Does it learn from each cycle - or repeat the same playbook?

Every collections cycle generates outcome data. Which channel got a response. Which script triggered payment. Which escalation path closed the account.

A real debt collection platform ingests these outcomes and continuously updates its routing decisions and scoring models. Month one, the platform is making decisions on baseline data. Month twelve, those decisions are materially sharper. Recovery rates compound.

A collections software product, by contrast, makes the same decisions every cycle on the same rules. Outcomes don't feed forward. The same routing mistakes are repeated indefinitely.

Ask the vendor: "How does the platform's behaviour at month 12 differ from its behaviour at month 1?" If the honest answer is "it doesn't, it just has more data" - you are buying a static system.

Question 4: Is compliance enforced in architecture - or in agent discretion?

The single most expensive failure mode for an Indian lender in 2026 is an RBI compliance incident. And the most common failure pattern is: the policy exists, the training happened, the script is correct - but an individual agent, on an individual call, on an individual day, did something the policy prohibited.

If the platform you are evaluating enforces compliance through agent training and quarterly memos, you have a policy. You do not have enforcement.

Real enforcement looks different:

Communication timing guardrails enforced at the platform layer - outreach is technically incapable of firing outside permitted hours.

Escalation sequences executed in order - every time - because the workflow engine, not the manager, holds the protocol.

Every borrower interaction is logged with timestamp, channel, content, and outcome - immutably.

Harassment pattern detection that flags and blocks risky communication before a complaint is filed.

Ask the vendor to show you the configuration screen where you set permitted communication hours. If those hours are stored as guidelines, you have policy. If they make the dialer technically incapable of firing outside them, you have architecture.

Question 5: What is the actual time-to-impact - 4 weeks or 6 months?

Most "platform" implementations take 4–9 months to go live. Custom integration, workflow configuration, data migration, UAT. During that period, your portfolio keeps aging into deeper DPD buckets, your team keeps running the old playbook, and the platform you bought is generating no impact.

A real CaaS (Collections as a Service) deployment goes live in 4–6 weeks. LMS/LOS integration via standard APIs. Workflow configuration in days, not months. The first measurable recovery uplift starts in week five.

The number to compare is not "implementation cost." It is "PAR exposure during the implementation period." A 6-month implementation on a ₹5,000 Cr book at 7% PAR is ₹350 Cr at risk during the build window. The fastest path to deployment is the cheapest path, even if the sticker price is higher.

Score your shortlist

Run every vendor on your shortlist through these five questions. Score each answer 0 to 2:

0 = no clear answer, or the answer reveals it's collections software dressed as a platform

1 = partial - the capability exists but is not the system's design point

2 = strong - clear evidence the platform was designed around this principle

A total of 8 or above is a genuine debt collection platform. Below 6, you are looking at collections software with newer marketing.

Get this evaluation right and the next 24 months of your portfolio look fundamentally different. Get it wrong and you will be back in this conversation, with a different vendor, in 18 months.